Overview
What is soundSDK?
soundSDK is a lightweight, offline sound recognition engine designed for embedded devices. It enables smart cameras, IoT devices, and other edge hardware to detect specific sounds such as baby crying, snoring, dog barking, cat sounds, glass breaking, alarm sounds, gunshots, knocking, and screaming.
Key Features
Ultra-Lightweight — INT8 models from 0.2–1 MB; latency and model tier configurable to your platform
Offline by Design — All inference runs locally, no network required
Cross-Platform — Supports ARM Linux, MIPS, and x86_64
Standard C API — Easy integration with any C/C++ codebase
Multi-Sound — Detect multiple sound types simultaneously
Architecture
The SDK consists of three layers:
1. Audio Capture Layer — Abstracted audio input (file, microphone, platform-specific)
2. Processing Layer — Mel-spectrogram extraction and sliding-window inference
3. Inference Engine Layer — Pluggable backends (Pure C FP32, INT8 quantized, NPU-accelerated)